At the time of its last public disclosure, Etched had just 35 employees (as of June 2024). That lean structure is typical of an early-stage hardware startup, but scaling to satisfy $1 billion in orders — let alone a $20 billion valuation — will require a massive expansion in engineering, sales, and support personnel, as well as a deep relationship with TSMC's advanced packaging and CoWoS capacity, which has been a supply-chain bottleneck across the industry.
The company has not disclosed revenue, gross margins, or a timeline for profitability. The $1 billion customer order book is a vote of confidence, but it does not yet translate into recognized sales. Investors in a $20 billion round would be betting that those orders convert, that margins on a custom chip exceed those on Nvidia GPUs, and that the hyperscale cloud providers — the ultimate buyers — are willing to diversify away from the CUDA ecosystem that has dominated AI training and inference for years.
A Competitive Landscape in Flux
Etched is hardly alone in chasing the transformer-inference opportunity. Large cloud operators are developing their own internal ASICs, and Nvidia itself is continually refreshing its inference offerings. Still, Etched's single-purpose architecture and its early traction with order commitments distinguish it from many earlier AI chip challengers that failed to gain commercial momentum.